Probabilistic fire spread forecast as a management tool in an operational setting

Abstract

Background: An approach to predict fire growth in an operational setting, with the potential to be used as a decision-support tool for fire management, is described and evaluated. The operational use of fire behaviour models has mostly followed a deterministic approach, however, the uncertainty associated with model predictions needs to be quantified and included in wildfire planning and decision-making process during fire suppression activities. We use FARSITE to simulate the growth of a large wildfire. Probabilistic simulations of fire spread are performed, accounting for the uncertainty of some model inputs and parameters. Deterministic simulations were performed for comparison. We also assess the degree to which fire spread modelling and satellite active fire data can be combined, to forecast fire spread during large wildfires events.

Results: Uncertainty was propagated through the FARSITE fire spread modelling system by randomly defining 100 different combinations of the independent input variables and parameters, and running the correspondent fire spread simulations in order to produce fire spread probability maps. Simulations were initialized with the reported ignition location and with satellite active fires. The probabilistic fire spread predictions show great potential to be used as a fire management tool in an operational setting, providing valuable information regarding the spatial-temporal distribution of burn probabilities. The advantage of probabilistic over deterministic simulations is clear when both are compared. Re-initializing simulations with satellite active fires did not improve simulations as expected.

Conclusion: This information can be useful to anticipate the growth of wildfires through the landscape with an associated probability of occurrence. The additional information regarding when, where and with what probability the fire might be in the next few hours can ultimately help minimize the negative environmental, social and economic impacts of these fires.

Keywords: FARSITE; Fire growth modelling; MODIS; Satellite active fires; Uncertainty assessment and propagation; VIIRS.

Fig. 1

Location of Portugal in Europe and location of the Tavira wildfire in Portugal

Fig. 2

Diagram relating modelling stages, input variables, model settings and simulation results

Fig. 3

Estimated probability of the adjustment factors for NFFL fuel models 1, 5, 6 and 9. The range represents the minimum and maximum estimated probability when considering all eight wildfires

Fig. 4

a Combined sampled point density of hourly wind direction and speed uncertainties; b combined sampled point density of daily relative humidity uncertainty and ROS adjustment factor for fuel model 6

Fig. 5

Ignition points and time-steps used in the simulations’ framework, based on reported and satellite information. a Ignition point(s) used to initialize simulations (day and time); b spatial and temporal information obtained from the Tavira wildfire reports and satellite active fires (day and time). See Table 1 for a more detailed description of the wildfire development and reference locations

Fig. 6

Simulations duration framework. Simulations ran sequentially from the ignition point(s) t₀, t₁, t₂, t₃ (start date) and between the defined time steps (end date)

Fig. 7

Probabilistic predictions (shaded colours) and deterministic simulations (blue line) of fire spread for the Tavira wildfire—t₀. Simulations initialized at the reported ignition point (start: day 18 at 13 h; durations: 25 (S3), 28 (S4), 31 (S5), 33 (S6), 35 (S7), 37 (S8), 48 (S10) and 57 (S11) hours, respectively) and fire front location at the specified hour (time step), obtained from both satellite and reported information

Fig. 8

Probabilistic predictions (shaded colours) and deterministic simulations (blue line) of fire spread for the Tavira wildfire—t₁. Simulations re-initialized with satellite active fires ignition points from the MODIS Aqua—VIIRS overpasses (start: day 19 at 3 h; durations: 11 (S12), 14 (S13), 17 (S14), 19 (S15), 21 (S16), 23 (S17), 34 (S19) and 43 (S20) hours, respectively) and fire front location at the specified hour (time step), obtained from both satellite and reported information

Fig. 9

Probabilistic predictions (shaded colours) and deterministic simulations (blue line) of fire spread for the Tavira wildfire—t₂. Simulations re-initialized with satellite active fires ignition points from the MODIS Aqua—VIIRS overpasses (start: day 19 at 14 h; durations: 3 (S21), 6 (S22), 8 (S23), 10 (S24), 12 (S25), 23 (S27) and 32 (S28) hours, respectively) and fire front location at the specified hour (time step), obtained from both satellite and reported information

Fig. 10

Probabilistic predictions (shaded colours) and deterministic simulations (blue line) of fire spread for the Tavira wildfire—t₃. Simulations initialized with satellite active fires ignition points from the MODIS Aqua—VIIRS overpasses (start: day 20 at 2 h; durantion: 11 (S29) and 20 (S31) hours, respectively) and fire front location at the specified hour (time step), obtained from both satellite and reported information

Fig. 11

Satellite active fires frequency distribution over the landscape main fuel types (at t₁, t₂ and t₃)

Fig. 12

Median burn probability at the fire front location at the specified hour (time step), for simulations initialized at t₀ and re-initialized at t₁, t₂ and t₃. The vertical lines represent the day-time of ignition